task_three_d__street__intersection_nearest_label

Summary

• Domain: three_d
• Scene id: street
• Scene package: street
• Supported query_id: single
• Answer type: option_letter
• Annotation schema: bbox

Program Contract

Program: select(label(candidate_street_objects, argmin(ground_distance_to_intersection_center))); scene=street; scope=intersection_nearest_label

Candidate set: the visible 3D objects, surfaces, room/street/warehouse structures, spatial anchors, markers, and labeled options inside the intersection_nearest_label objective scope. Operands: visible scene state and prompt-bound operands named by label, candidate_street_objects, argmin, ground_distance_to_intersection_center, street, intersection_nearest_label. Operation: evaluate select over the candidate set using the finalized 3D scene state, camera projection, object identities, spatial relations, counts, distances, or option-selection constraints encoded in the program expression; generation enforces a unique final answer. Output binding: answer uses the option_letter schema; generation binds a unique final answer. Annotation witnesses: annotation uses the bbox schema; the prompt/annotation contract defines the minimal visual witnesses. Query ids: single.

Reasoning Operations

Families: ranking, spatial_relations

Contract

The image shows a synthetic perspective 3D street intersection or T intersection with roads, sidewalks, crosswalk markings, unlettered street context, unlettered street-object candidates, and a below-scene text option panel. The street surface renders full-bleed: sidewalk ground fills the canvas and road strips continue to the visible image edges.

Each instance renders 4 unlettered candidate street objects. Exactly one candidate has the smallest finalized ground-plane distance from its object center to the intersection center, with a margin from the next-nearest candidate.

Street context can include buildings, storefronts, trees, shrubs, traffic lights, street signs, benches, and other unlettered objects. Context objects are visual context only and are excluded from answer candidates.

Annotation Contract

Annotation is the bounding box of the selected street object in the scene. Road markings, the intersection center, unlettered context objects, the option panel, and option text are not annotation.

Prompt And Trace

The prompt bundle is three_d_street_v1 under src/trace_tasks/resources/prompts/three_d/street/. The trace records camera pose, projection frame, scene variant, intersection layout, intersection center, full-bleed floor bounds, candidate ground positions, candidate distances to the intersection center, near-to-far label order, selected object id/type, projected object bboxes, and option-panel descriptors/bboxes.

Determinism

Generation is deterministic from instance_seed, explicit params, config defaults, prompt bundle, and code versions. Answers and annotation come from the same finalized 3D street scene trace.